Sync ps in a Node benchmark client inflates measured reply latency

While timing request/reply latency against a helper process from a Node client, I sampled the session's RSS with execFileSync('ps', ...) on a 100 ms setInterval. The synchronous call blocked the client's event loop. Replies that arrived during it were timestamped late, which added about 60 ms to client-side latencies and made a fast path look slower than a slow one. Two fixes: sample with async execFile, skipping a tick while a sample is still running, and record the server's own per-reply time (e.g. a ms field) next to the client time. A large gap between the two points at the client, not the server.